[stdlib] Make stdin/stdout/stderr Sendable on all platforms by JacobHearst · Pull Request #89891 · swiftlang/swift

JacobHearst · 2026-06-11T22:51:58Z

Resolves #75601.

Extends the existing OpenBSD shim functions in LibcOverlayShims.h to cover Linux and WASI, and re-exports them as nonisolated(unsafe) computed vars in the Glibc, Musl, Android, and WASILibc overlays.

finagolfin · 2026-06-14T16:48:37Z

@swift-ci test

finagolfin · 2026-06-14T16:49:00Z

@swift-ci test android

JacobHearst · 2026-06-15T12:21:33Z

Thanks for triggering those @finagolfin I'll take a look at the failures

finagolfin · 2026-06-15T14:25:01Z

@kateinoigakukun, @3405691582, and @al45tair, please take a look

@swift-ci test

finagolfin · 2026-06-15T14:25:32Z

@swift-ci test android

JacobHearst · 2026-06-15T17:17:11Z

I made the wrappers optional in my latest push because the first iteration of this change broke swift-log's usage of stderr and I (maybe incorrectly) thought that I should be able to achieve my goal without breaking existing usages. After the latest builds, I'm a bit confused.

swift-log force-unwraps stderr in this block (Logging.swift line 1254):

#elseif canImport(Glibc)
let systemStderr = Glibc.stderr!
let systemStdout = Glibc.stdout!
#elseif canImport(Musl)
let systemStderr = Musl.stderr!
let systemStdout = Musl.stdout!
#elseif canImport(WASILibc)
let systemStderr = WASILibc.stderr!
let systemStdout = WASILibc.stdout!

But the backtrace utils use stdout and stderr without force unwrapping (source).

I think what this means is that the original return type of the _swift_stdlib_stdxxx() functions was UnsafeMutableRawPointer! which would let callers use stderr or stderr!. Does this mean I should just make the wrappers return UnsafeMutablePointer<FILE>!?

finagolfin · 2026-06-16T04:12:23Z

@ahoppen, @grynspan, @weissi, would be good to get your feedback here.

weissi

thanks, this looks good to me.

In Swift 6 mode these were diagnosed as shared mutable global state (swiftlang#75601). Mark them nonisolated(unsafe) across the libc overlays. Glibc/Musl/Android/WASI define them as computed vars backed by the _swift_stdlib_std* shims; Platform.swift covers OpenBSD, FreeBSD/PS4, and Windows. They are OpaquePointer where FILE is opaque (Musl, WASI, Android, OpenBSD) and UnsafeMutablePointer<FILE> elsewhere. Adds the StdioSendable regression test.

JacobHearst · 2026-07-02T14:19:03Z

@finagolfin Could you trigger the tests again on the latest changes?

finagolfin · 2026-07-02T15:19:47Z

Hold on, Windows CI may be broken, will try in an hour or so if it isn't.

finagolfin · 2026-07-02T15:21:47Z

@kateinoigakukun and @al45tair, would be good to get your sign-off here, plus from any of @swiftlang/android-workgroup who have input.

finagolfin · 2026-07-02T17:26:10Z

Windows CI should be good now.

@swift-ci smoke test

finagolfin · 2026-07-03T10:22:06Z

@swift-ci test android

JacobHearst · 2026-07-09T12:52:36Z

Anything else needed from me to merge?

finagolfin · 2026-07-09T14:11:53Z

Just waiting to see if anyone else has input, think you've done your part. @ktoso, wdyt?

finagolfin · 2026-07-11T20:36:28Z

You'll need to rebase, some merge conflict.

ktoso

Long overdue, thanks for the PR! Looks good to me and enough other folks chimed in as well

ktoso · 2026-07-13T05:27:11Z

I took a stab at resolving the conflict, kicking off CI and we can merge if it passes.

ktoso · 2026-07-13T05:27:18Z

@swift-ci please test

swift-backtrace was using `setvbuf(stdout, nil, _IOLBF, 0)` which failed under `x86_64-swift-linux-musl` which broke after the recent changes in swiftlang#89891 We need to match the way those funcs are imported, so those variables must be `UnsafeMutablePointer<FILE>!` Also adds a test actually using those globals so we won't regress randomly. Resolves rdar://182172802

swift-backtrace was using `setvbuf(stdout, nil, _IOLBF, 0)` which failed under `x86_64-swift-linux-musl` which broke after the recent changes in #89891 We need to match the way those funcs are imported, so those variables must be `UnsafeMutablePointer<FILE>!` Also adds a test actually using those globals so we won't regress randomly. Resolves rdar://182172802

swift-backtrace was using `setvbuf(stdout, nil, _IOLBF, 0)` which failed under `x86_64-swift-linux-musl` after the recent changes in #89891. Musl's <stdio.h> forward-declares `struct _IO_FILE`, so Clang imports FILE as a nominal Swift type and every stdio function in the same overlay takes `UnsafeMutablePointer<FILE>`. Match that signature. Android (bionic) and WASI don't expose FILE as a Swift-nameable type at all, so their stdio functions come in typed with OpaquePointer. Keep those two overlays as OpaquePointer (initial fix attempt got that wrong -- flagged by finagolfin during review, ref swift-inspect String+Extensions.swift for the recommended `typealias File = OpaquePointer` bridging pattern under `#if os(Android)`). Also extends test/stdlib/StdioSendable.swift to actually pass the globals to stdio functions on the platforms where FILE is nominal (guards Android/WASI out). Resolves rdar://182172802

swift-backtrace was using `setvbuf(stdout, nil, _IOLBF, 0)` which failed under `x86_64-swift-linux-musl` after the recent changes in #89891 We need to match the way those funcs are imported, so those variables must be `UnsafeMutablePointer<FILE>!` Also adds a test actually using those globals so we won't regress randomly. Resolves rdar://182172802

…ssue/PR 90679) swift-backtrace was using `setvbuf(stdout, nil, _IOLBF, 0)` which failed under `x86_64-swift-linux-musl` after the recent changes in swiftlang/swift#89891 We need to match the way those funcs are imported, so those variables must be `UnsafeMutablePointer<FILE>!` Also adds a test actually using those globals so we won't regress randomly. Resolves rdar://182172802

) swift-backtrace was using `setvbuf(stdout, nil, _IOLBF, 0)` which failed under `x86_64-swift-linux-musl` after the recent changes in swiftlang#89891 We need to match the way those funcs are imported, so those variables must be `UnsafeMutablePointer<FILE>!` Also adds a test actually using those globals so we won't regress randomly. Resolves rdar://182172802

Recent pulls swiftlang#89891 and swiftlang#90679 tried to improve the async typing for Musl but broke the static linux SDK CI, so temporarily reverting.

Recent pulls #89891 and #90679 tried to improve the async typing for Musl but broke the static linux SDK CI, so temporarily reverting.

``` $ echo "@preconcurrency import var Foundation.stderr" | swiftly run +main-snapshot-2026-08-11 swiftc - -o /dev/null <stdin>:1:28: error: ambiguous name 'stderr' in module 'Foundation' 1 | @preconcurrency import var Foundation.stderr | `- error: ambiguous name 'stderr' in module 'Foundation' 2 | Glibc.stderr:1:32: note: found this candidate 1 | nonisolated(unsafe) public var stderr: UnsafeMutablePointer<FILE>! { get } | `- note: found this candidate /usr/include/stdio.h:151:14: note: found this candidate 149 | extern FILE *stdin; /* Standard input stream. */ 150 | extern FILE *stdout; /* Standard output stream. */ 151 | extern FILE *stderr; /* Standard error output stream. */ | `- note: found this candidate 152 | /* C89/C99 say they're macros. Make them happy. */ 153 | #define stdin stdin ``` swiftlang/swift#89891

…807) * PackageToJS: Derive WASI and shared memory support from wasm imports The SwiftBuild backend's build directory doesn't encode the target triple, so detect the traits derived from it by parsing the imports of the product binary instead. The imports are now parsed as a part of the packaging build graph. * Avoid `import var Foundation.stderr` ``` $ echo "@preconcurrency import var Foundation.stderr" | swiftly run +main-snapshot-2026-08-11 swiftc - -o /dev/null <stdin>:1:28: error: ambiguous name 'stderr' in module 'Foundation' 1 | @preconcurrency import var Foundation.stderr | `- error: ambiguous name 'stderr' in module 'Foundation' 2 | Glibc.stderr:1:32: note: found this candidate 1 | nonisolated(unsafe) public var stderr: UnsafeMutablePointer<FILE>! { get } | `- note: found this candidate /usr/include/stdio.h:151:14: note: found this candidate 149 | extern FILE *stdin; /* Standard input stream. */ 150 | extern FILE *stdout; /* Standard output stream. */ 151 | extern FILE *stderr; /* Standard error output stream. */ | `- note: found this candidate 152 | /* C89/C99 say they're macros. Make them happy. */ 153 | #define stdin stdin ``` swiftlang/swift#89891 * CI: Install Swift toolchain preserving its usr/bin layout Flattening the tarball into /usr/local makes SwiftPM derive a bogus toolchain root, so the swiftbuild build system cannot find swiftc. * PackageToJS: Record every import kind while parsing wasm imports Only memory imports were collected, so `WasmFeatures.isWASI` never saw `wasi_snapshot_preview1` and the generated node.js dropped the WASI import while still constructing a WASI instance. * PackageToJS: run per-target test runners under the swiftbuild build system `swift package js test` assumed the native build system's single combined `<Package>PackageTests` binary: it looked for one `.wasm`/`.xctest` under `.build/<config>/`, packaged it, and ran it. SwiftBuild produces no combined test binary. It emits one `<TestTarget>-test-runner.wasm` per test target under `.build/out/Products/`, plus an aggregate target that only orchestrates building them, so the old lookup failed with "Failed to find 'JavaScriptKitPackageTests.wasm'". (cherry picked from commit c12c7e9) * Install the JS event loop executor in async test targets Because SwiftBuild mode produces a linked executable for each test target. (cherry picked from commit 0a1f926) * BridgeJS: Skip type handle registration for unlinked modules `js test` generates the glue from every test target's skeletons, but the SwiftBuild build system links one binary per test target, so the eager registration hit a missing export. * PackageToJS: Share the linked JS modules with the aggregated glue The per-runner `bridge-js.js` copied to the base output directory for test preludes imports its JavaScript modules relative to itself, so copy the `bridge-js-modules` directory next to it. * BridgeJS: Require the imports object for static-only imported types An imported type looked up from `getImports` contributes its static methods too, but only a constructor marked the imports object as needed, so the glue referenced an undeclared `imports`. * PackageToJS: Generate each test bundle's glue from what its binary links in `js test` generated the glue from every test target's skeletons and handed the same copy to each SwiftBuild per-target runner, which described exports those binaries don't have. Package the aggregated glue once into the shared base directory, where preludes import it by a fixed path, and give each runner glue scoped to the target it links in. This supersedes the registration guard in the JS code generator.

JacobHearst requested a review from a team as a code owner June 11, 2026 22:51

JacobHearst force-pushed the sendable-stdio branch from 7992176 to 4417408 Compare June 15, 2026 13:08

3405691582 reviewed Jun 15, 2026

View reviewed changes

Comment thread stdlib/public/Platform/Glibc.swift.gyb Outdated

weissi approved these changes Jun 16, 2026

View reviewed changes

JacobHearst force-pushed the sendable-stdio branch from 4417408 to 2217c8a Compare June 29, 2026 21:19

JacobHearst force-pushed the sendable-stdio branch from 2217c8a to 73fee33 Compare June 29, 2026 22:48

3405691582 approved these changes Jun 30, 2026

View reviewed changes

al45tair approved these changes Jul 3, 2026

View reviewed changes

kateinoigakukun approved these changes Jul 8, 2026

View reviewed changes

ktoso approved these changes Jul 13, 2026

View reviewed changes

Merge branch 'main' into sendable-stdio

e4334e4

ktoso enabled auto-merge July 13, 2026 05:27

ktoso merged commit dc600dc into swiftlang:main Jul 13, 2026
5 of 6 checks passed

ktoso mentioned this pull request Jul 15, 2026

[NFC] Musl/Android/WASI stdio globals as UnsafeMutablePointer<FILE>! #90679

Merged

finagolfin mentioned this pull request Jul 29, 2026

[stdlib] Back out recent Musl changes for output/input streams #91064

Merged

finagolfin added a commit that referenced this pull request Jul 30, 2026

[stdlib] Back out recent Musl changes for output/input streams (#91064)

500e2bd

Recent pulls #89891 and #90679 tried to improve the async typing for Musl but broke the static linux SDK CI, so temporarily reverting.

finagolfin mentioned this pull request Aug 16, 2026

[stdlib] Back out recent linux/Glibc and Android changes for output/input streams #91525

Merged

finagolfin mentioned this pull request Aug 24, 2026

[SR-486] swiftc does not properly handle include of headers already imported from headers in other modules. #43103

Open

Conversation

JacobHearst commented Jun 11, 2026

Uh oh!

finagolfin commented Jun 14, 2026

Uh oh!

finagolfin commented Jun 14, 2026

Uh oh!

JacobHearst commented Jun 15, 2026

Uh oh!

finagolfin commented Jun 15, 2026

Uh oh!

finagolfin commented Jun 15, 2026

Uh oh!

Uh oh!

JacobHearst commented Jun 15, 2026

Uh oh!

finagolfin commented Jun 16, 2026

Uh oh!

weissi left a comment

Choose a reason for hiding this comment

Uh oh!

JacobHearst commented Jul 2, 2026

Uh oh!

finagolfin commented Jul 2, 2026

Uh oh!

finagolfin commented Jul 2, 2026

Uh oh!

finagolfin commented Jul 2, 2026

Uh oh!

finagolfin commented Jul 3, 2026

Uh oh!

JacobHearst commented Jul 9, 2026

Uh oh!

finagolfin commented Jul 9, 2026

Uh oh!

finagolfin commented Jul 11, 2026

Uh oh!

ktoso left a comment

Choose a reason for hiding this comment

Uh oh!

ktoso commented Jul 13, 2026

Uh oh!

ktoso commented Jul 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants